def test__db_cnode_status_and_two_jobs(self):
job1 = engine.prepare_job()
for node, status in [("P1", "up"), ("P2", "down"), ("P3", "down")]:
ns = models.CNodeStats(oq_job=job1,
node=node,
current_status=status)
ns.save(using="job_superv")
job2 = engine.prepare_job()
expected = {}
for node, status in [("Q2", "down"), ("Q3", "down")]:
ns = models.CNodeStats(oq_job=job2,
node=node,
current_status=status)
ns.save(using="job_superv")
expected[node] = ns
self.assertEqual(expected, monitor._db_cnode_status(job2.id))
def test_count_failed_nodes_with_a_node_that_went_offline(self):
# Result: 1 failed nodes
cs = models.CNodeStats(oq_job=self.job, node="N1", current_status="up")
self.db_mock.return_value = {"N1": cs}
self.live_mock.return_value = set()
actual = monitor.count_failed_nodes(self.job)
self.assertEqual(1, actual)
def test__db_cnode_status_and_wrong_job_id(self):
job = engine.prepare_job()
expected = {}
for node, status in [("O1", "up"), ("O2", "down"), ("O3", "down")]:
ns = models.CNodeStats(oq_job=job,
node=node,
current_status=status)
ns.save(using="job_superv")
self.assertEqual(expected, monitor._db_cnode_status(-1))
def test_count_failed_nodes_with_failures_before_calculation(self):
# Result: 1 node failure; this simulates the situation where a
# node has failed from the very beginning and never recovered i.e. it
# never took on any tasks. Only nodes that were functioning at some
# time during the calculation and *then* failed are counted.
n1 = models.CNodeStats(oq_job=self.job, node="N6", current_status="up")
n1.save(using="job_superv")
n2 = models.CNodeStats(oq_job=self.job,
node="N7",
current_status="down")
self.db_mock.return_value = {"N6": n1, "N7": n2}
self.live_mock.return_value = set()
actual = monitor.count_failed_nodes(self.job)
self.assertEqual(1, actual)
# The failed node has been updated to capture that.
n1 = models.CNodeStats.objects.get(id=n1.id)
self.assertEqual("down", n1.current_status)
self.assertEqual(1, n1.failures)
def test_count_failed_nodes_with_failures_during_calculation(self):
# Result: 2 node failures, please note that the function under test
# counts the total number of node failures that occurred during a
# calculation and *not* the number of currently failed nodes.
n1 = models.CNodeStats(oq_job=self.job, node="N3", current_status="up")
n2 = models.CNodeStats(oq_job=self.job,
node="N4",
current_status="down",
failures=1)
self.db_mock.return_value = {"N3": n1, "N4": n2}
self.live_mock.return_value = set(["N5"])
actual = monitor.count_failed_nodes(self.job)
self.assertEqual(2, actual)
# Please note also that the new node ("N5") was written to the
# database
[n3] = models.CNodeStats.objects.filter(oq_job=self.job, node="N5")
self.assertEqual("up", n3.current_status)
self.assertEqual(0, n3.failures)
def test_cnode_stats_failure_counter_with_up_down_transition(self):
# The failures counter is incremented in case of a
# up -> down transition
cs = models.CNodeStats(oq_job=self.job, node="N3", current_status="up")
cs.save(using="job_superv")
cs.current_status = "down"
cs.save(using="job_superv")
cs = models.CNodeStats.objects.get(id=cs.id)
self.assertEqual(1, cs.failures)
def test__db_cnode_status(self):
job = engine.prepare_job()
expected = {}
for node, status in [("N1", "up"), ("N2", "down"), ("N3", "down")]:
ns = models.CNodeStats(oq_job=job,
node=node,
current_status=status)
ns.save(using="job_superv")
expected[node] = ns
self.assertEqual(expected, monitor._db_cnode_status(job.id))
def test_cnode_stats_failure_counter_with_down_up_transition(self):
# The failures counter is only stepped in case of a
# up -> down transition
# and will remain unchanged here.
cs = models.CNodeStats(oq_job=self.job,
node="N6",
current_status="down")
cs.save(using="job_superv")
cs.current_status = "up"
cs.save(using="job_superv")
cs = models.CNodeStats.objects.get(id=cs.id)
self.assertEqual(0, cs.failures)
def count_failed_nodes(job):
"""Check compute nodes and return the total number of failures.
Please note that this function counts the total number of node
failures that occurred during a calculation and *not* the number
of currently failed nodes.
:param job: The :class:`openquake.db.models.OqJob` instance to use
:return: the number of failures i.e. how many nodes went from "up" to
"down" *at some time* during the calculation
"""
live_nodes = _live_cnode_status()
db_stats = _db_cnode_status(job)
if not live_nodes and not db_stats:
# No live compute nodes and nothing stored in the database; this will
# never work -> indicate failure
return -1
def set_status(node, status):
"""Update the status of the given node in the database."""
cs = db_stats[node]
cs.current_status = status
cs.save(using="job_superv")
# working nodes according to the database
dworking_nodes = set(cs.node for cs in db_stats.values()
if cs.current_status == "up" and cs.failures == 0)
# nodes that have failed at least once at some time during the calculation
dfailed_nodes = set(cs.node for cs in db_stats.values() if cs.failures > 0)
# Which working nodes stored in the db have gone bad/down?
total_failures = len(dfailed_nodes)
new_failed = dworking_nodes - live_nodes
for node in new_failed:
set_status(node, "down")
total_failures += 1
# Any entirely new nodes?
new_nodes = live_nodes - set(db_stats.keys())
for node in new_nodes:
cs = models.CNodeStats(oq_job=job, node=node, current_status="up")
cs.save(using="job_superv")
# Any nodes that came back after a failure?
for node in live_nodes.intersection(dfailed_nodes):
set_status(node, "up")
return total_failures
def test_cnode_stats_without_state_transition_and_same_timestamps(self):
# The `previous_ts` and `current_ts` time stamps are untouched
# if the compute node status did not change
cs = models.CNodeStats(oq_job=self.job,
node="N8",
current_status="down")
cs.save(using="job_superv")
old_current_ts = cs.current_ts
cs.node = "N8+1"
cs.save(using="job_superv")
cs = models.CNodeStats.objects.get(id=cs.id)
self.assertIs(None, cs.previous_ts)
self.assertEqual(old_current_ts, cs.current_ts)
def test_cnode_stats_with_state_transition_and_managed_timestamps(self):
# The `previous_ts` and `current_ts` time stamps are managed properly
# in case of a state transition
cs = models.CNodeStats(oq_job=self.job,
node="N7",
current_status="down")
cs.save(using="job_superv")
old_current_ts = cs.current_ts
cs.current_status = "up"
cs.save(using="job_superv")
cs = models.CNodeStats.objects.get(id=cs.id)
self.assertIsNot(None, cs.previous_ts)
self.assertEqual(old_current_ts, cs.previous_ts)
def test_count_failed_nodes_with_failed_and_recovered_node(self):
# Result: 1 node failure; the node failed and recovered. Its failures
# counter is unaffected by the recovery.
n1 = models.CNodeStats(oq_job=self.job, node="N8", current_status="up")
n1.save(using="job_superv")
self.assertEqual(0, n1.failures)
n1.current_status = "down"
n1.save(using="job_superv")
n1 = models.CNodeStats.objects.get(id=n1.id)
self.assertEqual(1, n1.failures)
self.db_mock.return_value = {"N8": n1}
self.live_mock.return_value = set(["N8"])
actual = monitor.count_failed_nodes(self.job)
self.assertEqual(1, actual)
# The failed node has been updated to capture that.
n1 = models.CNodeStats.objects.get(id=n1.id)
self.assertEqual("up", n1.current_status)
self.assertEqual(1, n1.failures)
def test_cnode_stats_with_correct_data(self):
# The db record is saved w/o triggering an exception
cs = models.CNodeStats(oq_job=self.job, node="N1", current_status="up")
cs.save(using="job_superv")